[LinuxBIOS] mmx problems on VIA EPIA-SP resolved

[Doug Bell]

We are working on VIA EPIA-SP with C3-2 processsor and have resolved
problems related to the use of the mmx as seen in the
auto.inc. The code was hanging unless we used -mno-mmx in Config.lb.

We then modified /src/cpu/via/model_centaur_init.c cpu_table[] to
include 
{X86_VENDOR_CENTAUR, 0698},   // VIA C3 Nehemiah

If anyone else is doing work on this board please let us know! Our code
isn't far enough along to check in.

Some questions remain:

1) Is the C3-2 supported yet? What's needed to add?

2) Which if any of the LinuxBIOS v2 via boards ( epia, epia-m ) in
subversion are building AND booting?  Has anyone done DRAM init using
SPD under v2?

Doug Bell - NXTV

[Li-Ta Lo]

On Thu, 2005-09-01 at 14:28 -0700, Doug Bell wrote:
> We are working on VIA EPIA-SP with C3-2 processsor and have resolved
> problems related to the use of the mmx as seen in the
> auto.inc. The code was hanging unless we used -mno-mmx in Config.lb.
>
> We then modified /src/cpu/via/model_centaur_init.c cpu_table[] to
> include 
> {X86_VENDOR_CENTAUR, 0698},   // VIA C3 Nehemiah
>
> If anyone else is doing work on this board please let us know! Our code
> isn't far enough along to check in.
>
> Some questions remain:
>
> Is the C3-2 supported yet? What's needed to add?
>
> Which if any of the LinuxBIOS v2 via boards ( epia, epia-m ) in
>
> subversion are building AND booting?  Has anyone done DRAM init using
> SPD under v2?

EPIA should use SPD in V2. It is not very stable for some unknown 
reason.

Doug Bell  writes:
> We are working on VIA EPIA-SP with C3-2 processsor and have resolved
> problems related to the use of the mmx as seen in the
> auto.inc. The code was hanging unless we used -mno-mmx in Config.lb.

Ok.  Likely an issue that romcc gets confused about which
instructions are part of the mmx,sse1, and sse2 extension.  As
the implementation filters by the registers used than by the
instructions.
> We then modified /src/cpu/via/model_centaur_init.c cpu_table[] to
> include 
> {X86_VENDOR_CENTAUR, 0698},   // VIA C3 Nehemiah

Sounds reasonable.
> If anyone else is doing work on this board please let us know! Our code
> isn't far enough along to check in.

A branch might work.

Eric